The Control of Reasoning Under Uncertainty
Abstract
The starting point of our research was to develop declarative representations of problem solving strategies and interpreters to select and execute appropriate strategies. One result was a 'strategy frame', a collection of fields or facets of strategies. These included the conditions under which strategies should be evoked, measures of progress, stopping conditions, resource requirements, and so on. We applied this result in a process control domain with some success, then began to search for a more realistic and demanding task. The Phoenix system, a realtime simulation of forest fire fighting in Yellowstone National Park, was an ideal testbed. In this context, our work on declarative representations of strategies followed two paths: We developed a declarative representation of the progress of plans, called envelopes. Envelopes provide early warning of plan failures, facilitate communication between agents, and help with monitoring and replanning in real time; and We began developing mathematical models of the architectures of agents that facilitate the design of agents for environments with different characteristics.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 14, 1991
- Accession Number
- ADA266308
Entities
People
- David M. Hart
- Paul R. Cohen
Organizations
- University of Massachusetts Amherst